Goldman Sachs: AI Bond Supply Lull Props Up Credit Spreads, $150 Billion in Chip Financing Poised to Launch
nashnova research
Goldman trader Jeffrey Papai argues investment-grade spread calm is not a risk signal but a supply vacuum — AI giants stopped issuing since September. When roughly $150 billion in chip financing returns, that calm breaks.
Rates whipsawed — why didn't credit spreads move?
Over the past six weeks the MOVE index — a gauge of rate volatility — breached 110. The Fed restarted hikes and French sovereign risk spread. Yet IG spreads widened by just 1 basis point.
Historical comparison: spreads widened 27 bp at the start of the 2015–16 hiking cycle and 15 bp in 2022. This cycle: near-zero reaction.
This means → Spreads held not because risk faded, but because the market's biggest issuers — AI giants — stopped selling new debt for over two months, temporarily removing supply-side pressure.
How rare is this supply gap?
Since September 2025 the market has gone two consecutive months without a single AI bond deal above $3 billion — the first time since the AI capex build began.
Before that pause, AI-related deals above $3 billion totalled 23 transactions worth $378 billion.
In plain terms = a group that had been borrowing heavily every month suddenly stopped. Fewer new bonds for sale meant existing bonds held their price — spreads tightened because there were no sellers, not because buyers grew more confident.
Why does the spread-to-yield ratio matter?
The IG index yield broke above 6% — only the third time since the financial crisis. In the six weeks after Jackson Hole, IG fund net inflows reached $12 billion.
Yet spreads as a share of total yield fell below 13% last month — a record low in both nominal and percentage terms. During the 2008 and 2020 stress peaks, that ratio exceeded 80%.
This means → The headline yield looks high, but almost all of it comes from the Treasury rate. The slice that actually compensates you for credit risk is thinner than ever — when risk arrives, the cushion is nearly gone.
What risk are AI heavyweights hiding inside the index?
Strip out Oracle, Microsoft, Google, Amazon, Meta, Nvidia, AMD, SuperMicro and Broadcom: the remaining index trades roughly 2–2.5 bp tighter than the full index. A year ago the gap was 0.5 bp wider.
Those nine names now account for over 6% of index market value (up from under 4%) and nearly 8% of index duration — a measure of interest-rate sensitivity — (up from under 5%).
Single-name signals are flashing: Broadcom 5-year CDS — the cost of insuring its debt — hit 136 bp, a record; Oracle hit 261 bp, also a record.
This reflects a shift: AI debt is no longer a small corner of the index — its credit pricing is dragging the entire IG market wider.
When $150 billion in chip financing returns, where does it hit?
Broadcom's "APCHIP 2.0" Anthropic TPU deal, SpaceX chip financing, the Broadcom–OpenAI "Jalapeno" project, and Oracle–Nvidia GPU financing are all in motion — roughly $150 billion combined, plus over $20 billion still outstanding from "APCHIP 1.0."
The key change: previous AI duration supply landed at the long end, but GPUs and TPUs depreciate fast, so future issuance will cluster at 7 years and shorter. Goldman estimates this segment will reach $275 billion by 2027 — 20%–25% of all short-end IG supply, up from about 10% today.
In plain terms = AI companies used to borrow long and compete with long-end buyers. Going forward they borrow medium-to-short. Regional banks, utilities and industrials rolling 5-year bonds will compete with OpenAI's GPU loans for the same pool of insurance money — significant pressure on the short-end credit curve.
What conditions does Papai attach to his trade call?
Papai sees the current setup as a buying opportunity; Q4 may deliver further spread tightening.
But he attaches a clear expiry: "until the 2027 AI supply wave approaches" — at that point the bill arrives and the logic breaks.
A second fragility: some insurers buying AI bonds draw confidence from the issuers' equity market caps. This means → if stock prices wobble, the credit thesis wobbles with them — a credit judgment built on equity valuations is itself a layer of risk.
市场有风险，内容仅供研究参考，不构成投资建议。